I'm interested in studying the role of various bank-specific and macroeconomic variables in determining stress in banks, using panel GMM estimation approach (motivated by persistence of risk in banks as per existing literature). My panel consists of 38 banks and 18 years. The variables are as follows.
Dependent Variable: Stress Scoreit or ln(Stress Scoreit)
Bank Specific Regressors: Pub_Dummy (1 for Public Sector bank, 0 for Private Sector Bank), Risk Leverage, GNPA, PCR, NIM, CorpLoan, Contingent Liabilities, Operating Efficiency, Size, RoA. (PCR and CorpLoan have been considered as predetermined variables, which are weakly exogenous)
Macroeconomic Variables: GDP Growth, Gsec Yield, Call Money Rate, Inflation, Exchange Rate, EPU Score
However on running two-step system GMM using -xtabond2, the following issues emerge.
1) All the macroeconomic variables are getting dropped due to collinearity (which is not the case, whlle using pooled OLS regression)
2) Coefficients of all bank specific variables except RoA and GNPA are coming out to be statstically insignificant at 95% confidence level (in pooled OLS regression, coefficients of all bank specific regressors except PCR, CorpLoan, Operating Efficiency and RoA are significant).
3) Hansen test p-value is coming out to be 1, which is potentially problematic (Roodman, 2009)
4) Difference-in-Hanesn statistic p-value, both for excluding group and difference, is coming out to be one, which I presume is problematic.
Kindly advise, where I'm going wrong, alternatively, how the estimation can be improved? I am new to panel GMM and have a very basic understanding of GMM estimation. The codes I have used are as follows.
1. Using StrsScore as dependent variable and other regressors/control variables at levels
2. Using StrsScore as dependent variable and other regressors/control variables at their first lag
I would greatly appreciate any suggestion.
Thanks
pankaj
Dependent Variable: Stress Scoreit or ln(Stress Scoreit)
Bank Specific Regressors: Pub_Dummy (1 for Public Sector bank, 0 for Private Sector Bank), Risk Leverage, GNPA, PCR, NIM, CorpLoan, Contingent Liabilities, Operating Efficiency, Size, RoA. (PCR and CorpLoan have been considered as predetermined variables, which are weakly exogenous)
Macroeconomic Variables: GDP Growth, Gsec Yield, Call Money Rate, Inflation, Exchange Rate, EPU Score
However on running two-step system GMM using -xtabond2, the following issues emerge.
1) All the macroeconomic variables are getting dropped due to collinearity (which is not the case, whlle using pooled OLS regression)
2) Coefficients of all bank specific variables except RoA and GNPA are coming out to be statstically insignificant at 95% confidence level (in pooled OLS regression, coefficients of all bank specific regressors except PCR, CorpLoan, Operating Efficiency and RoA are significant).
3) Hansen test p-value is coming out to be 1, which is potentially problematic (Roodman, 2009)
4) Difference-in-Hanesn statistic p-value, both for excluding group and difference, is coming out to be one, which I presume is problematic.
Kindly advise, where I'm going wrong, alternatively, how the estimation can be improved? I am new to panel GMM and have a very basic understanding of GMM estimation. The codes I have used are as follows.
1. Using StrsScore as dependent variable and other regressors/control variables at levels
Code:
. xtabond2 StrsScore L.StrsScore Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CM
> R EPUInd CPInfl ExcUSD i.Year,gmmstyle(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab L.OpEff L.Size L.R
> OA L.EPUInd, collapse) ivstyle(i.Year Pub_Dummy GDPGr GsecYld CMR CPInfl ExcUSD, equation(level)) twostep robust sma
> ll orthogonal
Favoring speed over space. To switch, type or click on mata: mata set matafavor space, perm.
GDPGr dropped due to collinearity
GsecYld dropped due to collinearity
CMR dropped due to collinearity
EPUInd dropped due to collinearity
CPInfl dropped due to collinearity
ExcUSD dropped due to collinearity
2005b.Year dropped due to collinearity
2019.Year dropped due to collinearity
Warning: Number of instruments may be large relative to number of observations.
Warning: Two-step estimated covariance matrix of moments is singular.
Using a generalized inverse to calculate optimal weighting matrix for two-step estimation.
Difference-in-Sargan/Hansen statistics may be negative.
Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: BankID Number of obs = 640
Time variable : Year Number of groups = 39
Number of instruments = 195 Obs per group: min = 14
F(28, 38) = 11.92 avg = 16.41
Prob > F = 0.000 max = 17
------------------------------------------------------------------------------
| Corrected
StrsScore | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
StrsScore |
L1. | 1.007106 .0290447 34.67 0.000 .948308 1.065904
|
Pub_Dummy | -38.60584 269.8552 -0.14 0.887 -584.8992 507.6875
CRAR | 6232.027 8747.618 0.71 0.481 -11476.6 23940.65
RiskLev | -2016.494 4204.734 -0.48 0.634 -10528.53 6495.545
GNPA | 1936.759 869.6131 2.23 0.032 176.3191 3697.199
PCR | 353.2481 441.8018 0.80 0.429 -541.1329 1247.629
NIM | 4897.158 27923.87 0.18 0.862 -51631.76 61426.08
CorpLoan | -4264.571 3052.743 -1.40 0.171 -10444.53 1915.383
ContLiab | -42.89541 52.97098 -0.81 0.423 -150.1296 64.33873
OpEff | -19178.99 26349.87 -0.73 0.471 -72521.51 34163.53
Size | 156.5269 101.4404 1.54 0.131 -48.82837 361.8821
ROA | 12847.05 4647.584 2.76 0.009 3438.505 22255.59
|
Year |
2006 | 154.515 241.632 0.64 0.526 -334.6435 643.6735
2007 | 109.6037 222.2718 0.49 0.625 -340.3621 559.5694
2008 | 21.27166 200.1552 0.11 0.916 -383.9214 426.4647
2009 | -30.94667 157.4384 -0.20 0.845 -349.664 287.7707
2010 | -159.2775 165.7282 -0.96 0.343 -494.7768 176.2217
2011 | -99.53624 189.7621 -0.52 0.603 -483.6896 284.6171
2012 | -67.23581 134.552 -0.50 0.620 -339.6221 205.1504
2013 | -19.35655 94.8411 -0.20 0.839 -211.3523 172.6392
2014 | 3.22885 113.5994 0.03 0.977 -226.7412 233.1989
2015 | -101.7981 181.5714 -0.56 0.578 -469.3702 265.774
2016 | -216.9991 80.92474 -2.68 0.011 -380.8227 -53.17556
2017 | -233.4362 95.06184 -2.46 0.019 -425.8789 -40.9936
2018 | -176.8919 65.95511 -2.68 0.011 -310.411 -43.37278
2020 | -245.7197 65.1777 -3.77 0.001 -377.6651 -113.7744
2021 | -174.6449 169.9768 -1.03 0.311 -518.7449 169.4552
2022 | -248.8271 105.2501 -2.36 0.023 -461.8948 -35.75938
|
_cons | 118.1285 626.1587 0.19 0.851 -1149.464 1385.72
------------------------------------------------------------------------------
Instruments for orthogonal deviations equation
GMM-type (missing=0, separate instruments for each period unless collapsed)
L(1/17).(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab
L.OpEff L.Size L.ROA L.EPUInd) collapsed
Instruments for levels equation
Standard
2005b.Year 2006.Year 2007.Year 2008.Year 2009.Year 2010.Year 2011.Year
2012.Year 2013.Year 2014.Year 2015.Year 2016.Year 2017.Year 2018.Year
2019.Year 2020.Year 2021.Year 2022.Year Pub_Dummy GDPGr GsecYld CMR CPInfl
ExcUSD
_cons
GMM-type (missing=0, separate instruments for each period unless collapsed)
D.(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab L.OpEff
L.Size L.ROA L.EPUInd) collapsed
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z = -1.98 Pr > z = 0.047
Arellano-Bond test for AR(2) in first differences: z = 0.14 Pr > z = 0.889
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(166) = 274.95 Prob > chi2 = 0.000
(Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(166) = 18.86 Prob > chi2 = 1.000
(Robust, but weakened by many instruments.)
Difference-in-Hansen tests of exogeneity of instrument subsets:
GMM instruments for levels
Hansen test excluding group: chi2(155) = 19.32 Prob > chi2 = 1.000
Difference (null H = exogenous): chi2(11) = -0.46 Prob > chi2 = 1.000
iv(2005b.Year 2006.Year 2007.Year 2008.Year 2009.Year 2010.Year 2011.Year 2012.Year 2013.Year 2014.Year 2015.Year 20
> 16.Year 2017.Year 2018.Year 2019.Year 2020.Year 2021.Year 2022.Year Pub_Dummy GDPGr GsecYld CMR CPInfl ExcUSD, eq(le
> vel))
Hansen test excluding group: chi2(160) = 20.72 Prob > chi2 = 1.000
Difference (null H = exogenous): chi2(6) = -1.85 Prob > chi2 = 1.000
2. Using StrsScore as dependent variable and other regressors/control variables at their first lag
Code:
. xtabond2 StrsScore L.StrsScore L.(Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld
> CMR EPUInd CPInfl ExcUSD) i.Year,gmmstyle(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab L.OpEff L.Size
> L.ROA L.EPUInd, collapse) ivstyle(i.Year Pub_Dummy GDPGr GsecYld CMR CPInfl ExcUSD, equation(level)) twostep robust
> small orthogonal
Favoring speed over space. To switch, type or click on mata: mata set matafavor space, perm.
L.GDPGr dropped due to collinearity
L.GsecYld dropped due to collinearity
L.CMR dropped due to collinearity
L.EPUInd dropped due to collinearity
L.CPInfl dropped due to collinearity
L.ExcUSD dropped due to collinearity
2005b.Year dropped due to collinearity
2011.Year dropped due to collinearity
Warning: Number of instruments may be large relative to number of observations.
Warning: Two-step estimated covariance matrix of moments is singular.
Using a generalized inverse to calculate optimal weighting matrix for two-step estimation.
Difference-in-Sargan/Hansen statistics may be negative.
Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: BankID Number of obs = 640
Time variable : Year Number of groups = 39
Number of instruments = 194 Obs per group: min = 14
F(28, 38) = 6.49 avg = 16.41
Prob > F = 0.000 max = 17
------------------------------------------------------------------------------
| Corrected
StrsScore | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
StrsScore |
L1. | 1.021737 .0384199 26.59 0.000 .9439603 1.099515
|
Pub_Dummy |
L1. | -808.5408 1598.502 -0.51 0.616 -4044.539 2427.458
|
CRAR |
L1. | -2378.348 5992.648 -0.40 0.694 -14509.83 9753.133
|
RiskLev |
L1. | 2362.036 3233.552 0.73 0.470 -4183.948 8908.021
|
GNPA |
L1. | 2116.038 4117.542 0.51 0.610 -6219.489 10451.57
|
PCR |
L1. | 410.4023 388.355 1.06 0.297 -375.7812 1196.586
|
NIM |
L1. | 9902.552 21569.14 0.46 0.649 -33761.89 53567
|
CorpLoan |
L1. | -3394.472 5579.028 -0.61 0.547 -14688.62 7899.68
|
ContLiab |
L1. | -92.75463 244.1143 -0.38 0.706 -586.9382 401.4289
|
OpEff |
L1. | -9366.682 24839.45 -0.38 0.708 -59651.52 40918.15
|
Size |
L1. | 289.8995 279.3452 1.04 0.306 -275.6053 855.4044
|
ROA |
L1. | 7861.62 11866.93 0.66 0.512 -16161.73 31884.97
|
Year |
2006 | 69.83238 241.5291 0.29 0.774 -419.1177 558.7824
2007 | 22.94654 235.2615 0.10 0.923 -453.3154 499.2085
2008 | 96.15762 150.0042 0.64 0.525 -207.5099 399.8251
2009 | 19.31248 105.8275 0.18 0.856 -194.924 233.549
2010 | -23.24291 121.6385 -0.19 0.849 -269.4872 223.0013
2012 | -134.4639 156.2665 -0.86 0.395 -450.8088 181.881
2013 | -49.08539 131.7116 -0.37 0.711 -315.7215 217.5507
2014 | -142.8812 122.3512 -1.17 0.250 -390.5683 104.8058
2015 | -289.6328 212.7753 -1.36 0.181 -720.3739 141.1083
2016 | -286.27 202.3132 -1.41 0.165 -695.8317 123.2917
2017 | -389.4327 317.6148 -1.23 0.228 -1032.41 253.5448
2018 | -395.2703 406.4983 -0.97 0.337 -1218.183 427.6424
2019 | -327.0297 412.6385 -0.79 0.433 -1162.373 508.3133
2020 | -532.6472 342.249 -1.56 0.128 -1225.494 160.1997
2021 | -391.5585 392.4236 -1.00 0.325 -1185.979 402.8615
2022 | -400.2929 337.4685 -1.19 0.243 -1083.462 282.8763
|
_cons | -460.5984 2259.37 -0.20 0.840 -5034.454 4113.258
------------------------------------------------------------------------------
Instruments for orthogonal deviations equation
GMM-type (missing=0, separate instruments for each period unless collapsed)
L(1/17).(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab
L.OpEff L.Size L.ROA L.EPUInd) collapsed
Instruments for levels equation
Standard
2005b.Year 2006.Year 2007.Year 2008.Year 2009.Year 2010.Year 2011.Year
2012.Year 2013.Year 2014.Year 2015.Year 2016.Year 2017.Year 2018.Year
2019.Year 2020.Year 2021.Year 2022.Year Pub_Dummy GDPGr GsecYld CMR CPInfl
ExcUSD
_cons
GMM-type (missing=0, separate instruments for each period unless collapsed)
D.(L.StrsScore PCR CorpLoan L.RiskLev L.GNPA L.NIM L.ContLiab L.OpEff
L.Size L.ROA L.EPUInd) collapsed
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z = -2.61 Pr > z = 0.009
Arellano-Bond test for AR(2) in first differences: z = -0.81 Pr > z = 0.421
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(165) = 281.75 Prob > chi2 = 0.000
(Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(165) = 14.81 Prob > chi2 = 1.000
(Robust, but weakened by many instruments.)
Difference-in-Hansen tests of exogeneity of instrument subsets:
GMM instruments for levels
Hansen test excluding group: chi2(154) = 8.68 Prob > chi2 = 1.000
Difference (null H = exogenous): chi2(11) = 6.13 Prob > chi2 = 0.864
iv(2005b.Year 2006.Year 2007.Year 2008.Year 2009.Year 2010.Year 2011.Year 2012.Year 2013.Year 2014.Year 2015.Year 20
> 16.Year 2017.Year 2018.Year 2019.Year 2020.Year 2021.Year 2022.Year Pub_Dummy GDPGr GsecYld CMR CPInfl ExcUSD, eq(le
> vel))
Hansen test excluding group: chi2(160) = 14.81 Prob > chi2 = 1.000
Difference (null H = exogenous): chi2(5) = -0.00 Prob > chi2 = 1.000
Thanks
pankaj
